TL;DR
This paper investigates the acoustic realization of the /d/ sound in spontaneous Spanish and Catalan speech, revealing that its weakening (lenition) is a gradual process, not a strict binary switch between stop and fricative. The study introduces novel acoustic metrics to quantify this lenition, showing that the preceding sound significantly influences how "weakened" the /d/ becomes.
What Problem Does This Paper Address?
In many Romance languages, particularly those in the Iberian Peninsula like Spanish and Catalan, voiced plosive consonants (/b/, /d/, /g/) undergo a process called lenition, often manifesting as a transition from a stop sound to a fricative or approximant. Traditionally, this has been described as a categorical phenomenon: the sound is either a stop or a fricative, depending on its phonetic environment (e.g., word-initial vs. intervocalic). However, previous research has hinted at a more nuanced, continuous variation in the degree of constriction. This paper aims to explore this variation acoustically, specifically for the /d/ sound, and to determine if the process is truly a gradient one or if it can be better explained by a bimodal distribution (suggesting two distinct allophones). The authors also investigate the role of the preceding phonetic segment as a predictor of this lenition.
Methods
The researchers developed three novel acoustic metrics to quantify the degree of constriction of the /d/ sound:
- Intensity Difference (ID): This metric measures the difference in acoustic intensity between the minimum point during the /d/ sound and the maximum point in the following vowel. A smaller difference suggests a more open constriction (more lenition).
- Spectral Tilt (ST): This metric calculates the difference in acoustic energy between a low-frequency band (50-500 Hz) and a high-energy band (500-5000 Hz) within the /d/ sound. More energy in the higher band (indicating a more open constriction) leads to a smaller difference between the bands.
- Maximum Velocity (MV): This metric analyzes the first difference of the intensity curve over time, specifically looking at the maximum velocity between the intensity minimum of the /d/ and the maximum of the following vowel. More constricted consonants are expected to have more abrupt transitions, resulting in higher MV values.
These metrics were applied to manually segmented speech data. Statistical analysis was performed using mixed-effects regression models, with individual speakers and lexical items treated as random factors.
Datasets
The study utilized two corpora of spontaneous, interactive speech:
- Spanish: 241 /d/ tokens from 20 speakers.
- Catalan: 264 /d/ tokens from 20 speakers.
These datasets were collected from a bilingual speech community on the island of Majorca, Spain. The data were recorded using head-mounted microphones and solid-state recorders at a sampling rate of 44.1 kHz and 16-bit depth.
Key Insights and Ideas
- Lenition is a Continuum: The acoustic analyses consistently revealed unimodal distributions for all three metrics in both Spanish and Catalan. This strongly suggests that the lenition of /d/ is a gradual process, not a strict binary distinction between stop and fricative allophones.
- Preceding Segment Matters: The nature of the sound immediately preceding the /d/ is a significant predictor of its degree of constriction. Consonants generally lead to more constricted /d/ sounds compared to vowels.
- Vowel Height Influences Lenition: In Spanish, the height of the preceding vowel plays a crucial role. Lower vowels lead to more weakened /d/ sounds than higher vowels. A similar trend was observed in Catalan, though less robustly.
- Language Differences are Subtle: While there are some differences in lenition patterns between Spanish and Catalan, particularly in the spectral tilt of the /d/, the overall constriction degree doesn't differ dramatically across the two languages when considering all metrics.
Key Results
- Spanish:
- Low and mid-vowel contexts resulted in more weakened /d/ tokens than high-vowel contexts for the Intensity Difference (ID) metric.
- Similar patterns were observed for Spectral Tilt (ST), with low and mid vowels preceding more weakened /d/s.
- Maximum Velocity (MV) showed that low vowels, nasals, and sibilants triggered more weakened /d/s than high vowels, while nasals and sibilants triggered more constricted /d/s than high vowels.
- Catalan:
- The Intensity Difference (ID) showed that /d/ tokens preceded by schwa, nasals, or sibilants differed from those preceded by high vowels. Specifically, schwa preceded more weakened /d/s, while nasals and sibilants preceded more constricted /d/s.
- Spectral Tilt (ST) showed a difference between sibilant and high-vowel contexts, with sibilants leading to more constricted /d/s.
- Maximum Velocity (MV) showed differences between consonantal contexts (/n/, /s/) and vowel contexts, but no clear distinction between the two vowel contexts or the two consonant contexts.
- Cross-linguistic Comparison: The Spectral Tilt (ST) metric revealed a significant difference between Catalan and Spanish /d/, with Catalan /d/ tokens showing lower (more negative) values, suggesting potentially more constriction. However, intensity-based measures (MV and ID) did not show significant differences between the languages.
Limitations and Future Work
- Limited Contexts: For both languages, the analysis was restricted to specific preceding segments due to insufficient token counts in other environments. For Catalan, the analysis of vowel contexts was limited to high vowels and schwa.
- Metric Correlation: While the three acoustic metrics capture aspects of constriction, they are only modestly correlated with each other, suggesting they capture different facets of the phenomenon.
- Future Work: The authors suggest that further research could explore the interplay of other factors like speaking rate and lexical stress more deeply. They also highlight the need for more robust acoustic-to-articulatory mappings to fully understand the nuances of lenition. The subtle differences observed between Catalan and Spanish /d/ warrant further investigation with larger datasets and potentially different phonetic environments.
Summary updated